New, Free 165-Language Dictionary Offers Promise for Developing Speech Recognition Technology in Hundreds of Languages
New open-source software created by linguists at The Graduate Center, CUNY may speed the development of speech technologies such as recognizers and synthesizers in new languages.
New open-source software created by linguists at The Graduate Center, CUNY may speed the development of speech technologies such as recognizers and synthesizers (think Siri) in new languages. The software, called WikiPron, has already generated a database of pronunciations in 165 languages, from Adyghe to Zulu. With the database and the mining software, programmers can build and evaluate speech technologies in languages that have received little attention in the past.
“Technologies like virtual assistants and spoken driving directions need to know how to pronounce words, including ones the system may have not seen before,” said lead researcher Kyle Gorman, a professor of linguistics at The Graduate Center, CUNY who began his career at Google. “WikiPron helps us increase our coverage of the world's languages using data produced by volunteers that write and edit Wiktionary, an open-access platform.”
Wiktionary is a free, open-access website that crowdsources information on pronunciations in more than 900 languages. By mining Wiktionary’s data, WikiPron generated pronunciations for 1.7 million words in 165 languages. WikiPron can also be used to predict the pronunciations of new words. This kind of artificial intelligence is invaluable for developing so-called smart products, or assistive text-to-speech technologies.
WikiPron stands apart from other speech pronunciation software for being open source and freely available, and it encompasses more languages than similar software.
The research team is also employing WikiPron for a collaborative online project that encourages developers to create artificial intelligence tools to predict the pronunciation of unfamiliar words in 15 languages.
Gorman and his fellow researchers describe their work in a paper published by the Association for Computational Linguistics. Co-authors include Graduate Center, CUNY Ph.D. students Lucas Ashby and Yeonju Lee-Sikka and master’s students Elizabeth Garza, Alan Wong, and Sean Miller.